Tilburg University

The challenge of retaining customers acquired with free trials

Datta, H.; Foubert, B.; van Heerde, H.J.

Published in:Journal of Marketing Research

DOI:10.1509/jmr.12.0160

Publication date:2015

Document VersionPeer reviewed version

Link to publication in Tilburg University Research Portal

Citation for published version (APA):Datta, H., Foubert, B., & van Heerde, H. J. (2015). The challenge of retaining customers acquired with free trials.Journal of Marketing Research, 52(2), 217-234. https://doi.org/10.1509/jmr.12.0160

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal

Take down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Download date: 14. Mar. 2022

The Challenge of Retaining Customers Acquired with Free Trials

Hannes Datta

Bram Foubert

Harald J. van Heerde*

*Hannes Datta is Assistant Professor at the Department of Marketing, Tilburg University, P.O. Box 90153, 5000 LE

Tilburg, The Netherlands, h.datta@tilburguniversity.nl. Bram Foubert is Assistant Professor at the Department of

Marketing and Supply Chain Management, Maastricht University, P.O. Box 616, 6200 MD Maastricht, The

Netherlands, b.foubert@maastrichtuniversity.nl. Harald J. van Heerde is Research Professor of Marketing, Massey

Business School,, Massey University, Private Bag 102904, Auckland 0745, New Zealand, and Extramural Fellow at

CentER, Tilburg University, The Netherlands, heerde@massey.ac.nz.

Acknowledgements. The authors gratefully acknowledge Marnik Dekimpe, Aurélie Lemmens, and Sungho Park, and

wish to thank Johannes Boegershausen, Kelly Geyskens, Caroline Goukens, Anne Klesse, and Arjen van Lin for

helpful comments on an earlier draft. The first author acknowledges the financial support of the Graduate School of

Business and Economics of Maastricht University, and the Netherlands Organization for Scientific Research (NWO

Vici Grant 453-09-004). This work was carried out on the Dutch national e-infrastructure with the support of SURF

Foundation.

1

The Challenge of Retaining Customers Acquired with Free Trials

ABSTRACT

Many service firms acquire customers by offering free-trial promotions. A crucial

challenge is to retain customers acquired with these free trials. To address this challenge, firms

need to understand how free-trial customers differ from regular customers in terms of their

decision making to retain the service. This article conceptualizes how a customer’s retention

decision is driven by marketing communication and usage behavior, and develops hypotheses

about the impact of free-trial acquisition on this process. To test the hypotheses, the authors

model a customer’s retention and usage decisions, distinguishing between usage of a flat-rate

service and usage of a pay-per-use service. The model allows for unobserved heterogeneity and

corrects for selection effects and endogeneity. Based on household panel data from a digital TV

service, the authors find systematic behavioral differences which make the average customer

lifetime value (CLV) of free-trial customers 59% lower than that of regular customers. However,

free-trial customers are more responsive to marketing communication and usage rates, which

offers opportunities to target marketing efforts and enhance retention rates, CLV, and customer

equity.

KEYWORDS: Free Trials, Customer Retention, Usage Behavior, Customer Lifetime Value,

Customer Equity, Acquisition Mode, Selection Effects, Endogeneity, Econometrics

2

A very popular way to acquire new customers, especially among service providers, is to

offer free-trial promotions. Customers on a free trial are allowed to try the service for a limited

amount of time at no charge. Well-known examples are the free trials offered by mobile

telephone operators (e.g., AT&T in the US), video streaming websites (e.g., Netflix in the US),

and digital TV providers (e.g., Sky television in Australia and New Zealand). While these free

trials may be popular with consumers, a crucial challenge to firms is to retain customers who

have been acquired with a free trial. To address this challenge, firms need to understand whether

customers attracted with free trials are systematically different from other customers. In this

article, we argue that free-trial acquisition may affect the nature of a customer’s relationship with

the service provider and, as a consequence, influence usage and retention behavior, consumers’

responsiveness to marketing activities, and ultimately customer lifetime value (CLV).

An emerging body of research has shown that the conditions under which customers are

acquired have implications for subsequent consumer behavior (e.g., Reinartz, Thomas, and

Kumar 2005; Schweidel, Fader, and Bradlow 2008). A first group of studies documents the role

of the sales channel through which customers have been attracted (e.g., Steffes, Murthi, and Rao

2011). For example, Verhoef and Donkers (2005) find that acquisition through the Internet leads

to higher retention rates than acquisition through direct mail or direct-response commercials. A

second set of papers addresses the impact of customer referral (Chan, Wu, and Xie 2011;

Schmitt, Skiera, and Van den Bulte 2011). Villanueva, Yoo, and Hanssens (2008), for instance,

show that customers acquired through word-of-mouth referral have longer lifetimes.

A third stream of research, in which we position our own work, examines the effects of

the price structure or promotional conditions under which customers were acquired. Iyengar et

al. (2011) find that customers of a telecommunication company who were acquired under a two-

3

part tariff structure have lower usage and retention rates than customers charged on a pay-per-use

basis.1 Lewis (2006) shows that acquisition discounts lead to lower retention rates, while

Anderson and Simester’s (2004) results indicate that promotionally acquired customers choose

cheaper products but buy more. Table 1 summarizes the relevant research.

[PLEASE INSERT TABLE 1 ABOUT HERE]

This article contributes to the literature in three ways. First, while previous work has

examined the impact of promotional customer acquisition on subsequent behavior, the effects of

free-trial promotion have remained largely unaddressed. A free trial involves a distinct type of

sales promotion which enables consumers to start using a service without a financial obligation

and to revise their initial adoption decision if they are not satisfied. A free trial thus allows

consumers to engage in a low-commitment relationship with the firm (Dwyer, Schurr, and Oh

1987). Relying on buyer-seller relationship theory, we will argue that this may lead to systematic

differences in behavior between free-trial and regular customers.

Although some studies examine the effects of free trials and sampling, most look at

aggregate sales (e.g., Heiman et al. 2001; Jain, Mahajan, and Muller 1995; Pauwels and Weiss

2008) or focus on immediate purchase effects (Scott 1976). Gedenk and Neslin (1999), who do

study individual customer behavior, find that sampling in the mineral water category reinforces

choice probabilities after the promotion. However, the authors do not examine retention, as this

is not relevant for FMCGs. Bawa and Shoemaker (2004) show that free samples attract new

buyers who may remain customers in subsequent periods. Yet, it is unclear whether the retention

rates of customers attracted with a sample differ from those of regular customers.

As a second contribution, we extend insights on the role of usage behavior in the

customer value generation process. Specifically, usage intensity can be an important driver of

4

retention because it reminds customers about the personal value of the service (e.g., Bolton and

Lemon 1999; Prins, Verhoef, and Franses 2009). This paper adds to these insights by examining

how free-trial acquisition influences the relationship between usage and retention. In particular,

if this relationship turns out to be particularly strong for customers acquired via free trial, it is in

the firm’s interest to encourage usage especially among these customers. An important

consideration in this respect is that many services involve two types of usage: flat-rate usage

(e.g., a regular TV subscription) and pay-per-use consumption (e.g., video-on-demand). While

both types of usage drive retention, usage of a pay-per-use service is also a direct source of

revenues (e.g., Danaher 2002; Iyengar et al., 2011). Therefore, we distinguish between flat-rate

and pay-per-use service components and examine the role of usage, not only as an antecedent of

retention but also as a direct component of CLV.

Third, a crucial yet unexplored question is whether acquisition mode affects customers’

responsiveness to the firm’s marketing communication efforts. Therefore, we evaluate the

differences in marketing responsiveness between free-trial and regular customers. Specifically,

we consider customers’ reaction to direct marketing and traditional advertising because of the

growing interest in marketing communication as a way to actively manage customers’ tenure

(e.g., Polo, Sese, and Verhoef 2011; Reinartz, Thomas, and Kumar 2005). If free-trial and

regular customers respond differently to direct marketing and advertising, firms may decide to

target retention efforts to the most receptive group to reduce churn.

In sum, we investigate whether free-trial acquisition influences retention behavior and

CLV, and how it moderates the extent to which retention is driven by service usage (flat-rate and

pay-per-use) and marketing communication. We develop econometric models for customers’

usage and retention decisions, accounting for unobserved heterogeneity and endogenous

5

marketing instruments. Importantly, because we are interested in the impact of free-trial

acquisition on a customer’s behavior, we correct for selection effects. In particular, free trials

may attract consumers with, a priori, lower valuations of the service (e.g., Lewis 2006). While

previous research on the role of acquisition mode typically ignores selection effects (see Table

1), we consider two alternative approaches to address these effects, i.e., explicit modeling of the

selection process (e.g., Thomas 2001) and matching (e.g., Gensler, Leeflang, and Skiera 2012).

Based on household panel data for more than 16,000 customers of a large European

digital TV provider, we find that free-trial customers have lower retention rates and use the

firm’s flat-rate service less intensively than regular customers. As a result, their CLV is, on

average, 59% lower than that of regular customers. However, free-trial customers are more

responsive to marketing communication and more likely to rely on their usage behavior when

deciding whether or not to retain the service. These findings offer managers opportunities to

better target their marketing efforts and improve retention rates, CLV, and customer equity.

CONCEPTUAL FRAMEWORK AND HYPOTHESES

Figure 1 presents the conceptual framework for this research. The core consists of a

customer’s usage and retention decisions, which are influenced by the acquisition mode, i.e.,

free-trial versus regular acquisition.

[PLEASE INSERT FIGURE 1 ABOUT HERE]

Core Decision Process

The core decision process involves two types of periodic (e.g., monthly) decisions. Every

period, consumers decide how intensively to use the service and whether to retain it or not.

Service usage. We distinguish between two types of service usage that are common for

subscription services: (1) usage of a flat-rate service, which is included in the subscription

6

charges, and (2) usage of a pay-per-use service, for which consumers pay per unit of

consumption. While both types of usage may foster retention, consumption of the pay-per-use

service, in addition, directly generates revenue.

Service retention. Every period, consumers decide whether to retain the service or not.

We distinguish between two sets of drivers for this decision. First, consumers rely on their usage

intensity for the flat-rate and pay-per-use component to assess the utility of retaining the service

(Bolton and Lemon 1999). As a result, a high usage intensity will stimulate retention, whereas a

low usage rate may lead to disadoption (Lemon, White, and Winer 2002). Note that, compared to

usage of the pay-per-use service, flat-rate usage may be more consequential for customers’

evaluation of the service subscription, because it is included in the fixed periodical fee (Bolton

and Lemon 1999). Second, a consumer’s retention decision is also influenced by marketing

communication (e.g., Blattberg, Malthouse, and Neslin 2009). Specifically, direct marketing and

advertising remind customers of the benefits of the service or directly persuade them to retain it.

If consumers retain the service in the current period, they go through the same usage and

retention decision process in the following period. As indicated by the dashed lines, this repeated

decision process drives customer lifetime value (CLV). In particular, the periodic retention

decisions generate a stream of fixed subscription fees (which cover flat-rate usage) while usage

of the pay-per-use service generates additional revenue.

Differences between Free-Trial and Regular Customers

Central to our study is the expectation that the decision process to use and retain the

service differs between free-trial and regular customers. Our hypotheses build on buyer-seller

relationship theory, which posits that customer behavior depends on the nature of the relationship

between customer and firm (Dwyer, Schurr, and Oh 1987; Johnson and Selnes 2004).

7

Baseline retention. Drawing on relationship theory, we expect free-trial customers to

churn sooner than regular customers. Whereas the anticipated longevity of a regular contract

encourages customers to immediately commit to the firm, subscription to a free trial resembles a

discrete transaction which merely increases a consumer’s awareness of the firm and facilitates

relationship exploration (Dwyer, Schurr, and Oh 1987; Johnson and Selnes 2004). According to

self-perception theory, free-trial customers may make post-hoc inferences about the reasons for

their behavior and hence attribute their adoption decision to the availability of a free trial rather

than to a strong commitment to the company (e.g., Dodson, Tybout, and Sternthal 1978; Gedenk

and Neslin 1999). In other words, a free trial slows down the relationship formation process

(Palmatier et al. 2013). Importantly, even after the free-trial period has expired, the firm’s

relationship with free-trial customers likely remains more fragile than that with regular

customers. In particular, research by Gilbert and Ebert (2002) and Gilbert et al. (1998) indicates

that consumers who receive the opportunity to first evaluate a product or service are more critical

than when they immediately commit to the firm, a tendency that persists after the evaluation

period. That is, the critical reflections generated during exploration of a relationship remain

active, even when the customer moves to a closer relationship level. Hence, we hypothesize that:

H1 Free-trial customers have a lower retention rate than regular customers even after

the free trial expires.

Impact of usage on retention. Because customers attracted with a free trial arguably have

a less developed relationship with the firm than regular customers, they may be more uncertain

about the service benefits (Johnson and Selnes 2004). One major factor that informs consumers

about the personal value of the service and thus helps to resolve the uncertainty, is their own

usage behavior (e.g., Bolton and Lemon 1999). A customer may wonder: Do I use the service

8

enough to stay subscribed? We expect that, to overcome their uncertainty, free-trial customers

are more inclined than regular customers to assess the service’s value on the basis of their flat-

rate and pay-per-use consumption. Regular customers, who are more committed to the firm, are

less likely to base their retention decision on usage intensity. Hence, we expect that:

H2a The impact of usage of a flat-rate service on retention is greater for free-trial

customers than for regular customers.

H2b The impact of usage of a pay-per-use service on retention is greater for free-trial

customers than for regular customers.

Impact of marketing communication on retention. We also argue that the firm’s

marketing communication, in the form of direct marketing and advertising, will be more

important to free-trial than to regular customers. Marketing communication provides free-trial

customers with information that can compensate for their relatively high uncertainty (Mitchell

and Olson 1981). Regular customers, in contrast, may be less susceptible to external information

because of confidence in their level of expertise (Brucks 1985; Hoch and Deighton 1989). In line

with these principles, Johnson and Selnes (2004) postulate that it is easier to boost commitment

among customers in a lower-level relationship with the firm than among already dedicated

customers. As a result, we expect that free-trial customers are more responsive to the firm’s

direct-marketing and advertising efforts than regular customers:

H3a The impact of direct marketing on retention is greater for free-trial customers than

for regular customers.

H3b The impact of advertising on retention is greater for free-trial customers than for

regular customers.

Baseline usage. Free-trial acquisition may also affect customers’ usage intensity. On the

9

one hand, free-trial customers are less committed to the firm and less convinced of the service

benefits, such that they may have lower usage rates than regular customers. In fact, usage is one

of the most tangible reflections of engagement with the firm (Van Doorn et al. 2010). This holds

true in particular for usage of the flat-rate service, which is the main object of the contractual

relationship (Bolton and Lemon 1999). On the other hand, exactly because free-trial customers’

relationship with the firm is more exploratory in nature (Dwyer, Schurr, and Oh 1987; Gilbert et

al. 1998), they may use the service more frequently to become more certain about the service’s

benefits. These opposing principles do not allow us to develop unidirectional expectations

regarding the impact of free-trial acquisition on usage of the flat-rate and pay-per-use services.

DATA

Study Context

We test the hypotheses based on a household panel data set from a large European

interactive TV (iTV) provider. iTV is a technology that enables customers to interact with their

TV, for example, by browsing an electronic program guide or watching video-on-demand

(VOD). Furthermore, iTV offers enhanced image quality over regular TV. To use the iTV

service, customers need a broadband DSL Internet connection from the same company and a

set-top box which decodes the digital signal. The focal company is the only provider of digital

TV via DSL and, at the end of the observation period, had a market share of 31% in the digital

TV market. Its main competitor, which offers digital TV via cable, had a market share of 40%.2

Under regular conditions, the company’s customers formally commit to a 12-month

subscription period. They have the possibility to cancel the service earlier, in which case they

pay a penalty (€50, plus €6 for every remaining month until the end of the contractual period).

After the first twelve months, the contract is automatically renewed but can be terminated each

10

month without penalty. Customers are charged a one-time setup fee for hardware and activation

(on average €16.24) and pay for service usage according to a two-part tariff structure (e.g.,

Ascarza, Lambrecht, and Vilcassim 2012; Iyengar et al. 2011). Specifically, the fixed monthly

subscription fee of €15.95 covers unlimited usage of the basic iTV service (€9.95) and rent of the

set-top box (€6). Additionally, customers can make use of a video-on-demand (VOD) service,

for which they are charged on a pay-per-use basis. They can select VODs from an electronic

catalogue containing movies, live concerts, and soccer games. VOD rental for 24 hours costs

around €3, with some limited variation in price due to differences in genre and length.

The company’s acquisition strategy offers a unique setting to study the impact of

free-trial acquisition. For a period of 10 months (months 10 till 19 after launch of the service),

the company offered free trials parallel to its regular subscriptions. Adoption of the free trial (as

opposed to the regular subscription) is largely driven by consumers’ awareness of the ongoing

free-trial promotion, which was mainly promoted via direct marketing. Customers acquired with

a free trial did not pay setup costs and were not charged monthly subscription fees for the usage

of the flat-rate service during a three-month period. VOD usage, however, was not free of

charge. Free-trial customers could revise their adoption decision by returning the set-top box to

one of the company stores before the end of the trial period, without paying a penalty. If the

product was not returned by the end of the three-month trial period, the subscription was

converted into a paid one, such that the next nine months were considered part of a regular

contract.

Data set

From the initial sample of close to 21,000 customers who adopted iTV when both the free

trial and regular subscription were available, we retain a subset based on several criteria.

11

Specifically, we eliminate customers who had missing socio-demographics, were employees of

the focal company, or did not speak the local language (and thus could not understand the

advertising and direct-marketing messages). We thus retain 16,512 customers of which 12,612

(76%) were acquired with free trials and the balance of 3,900 (24%) signed a regular contract.

We observe customers’ retention and usage behavior until two years after launch of the

service. 6,079 (48%) of the free-trial customers churn before the end of the observation period,

while of the regular customers only 1,327 (34%) do so. Furthermore, the data set includes two

types of usage: (1) flat-rate usage of the basic interactive TV service, which is measured by a

customer’s monthly number of channel zaps,3 and (2) usage of the VOD service, for which we

use the monthly number of VODs the customer has watched. Compared to regular customers,

free-trial customers have an average usage intensity that is 11% lower for the flat-rate service

(169 versus 189 zaps per month), but 26% higher for the video-on-demand service (.73 versus

.58 VODs per month). However, these retention and usage measures are only indicative, because

they do not account for selection effects or the impact of marketing activities.

The company uses two types of marketing communication: direct marketing and mass

advertising. We operationalize direct marketing as the monthly number of direct-marketing

contacts with a given customer (via phone, e-mail, or regular mail). On average, free-trial and

regular customers are contacted .34 and .16 times per month, respectively. In the analyses, we

account for these systematic differences in contact frequency. In addition, the data set includes a

measure for the company’s spending on mass advertising (via TV, print media, radio, and the

Internet). In particular, this variable quantifies the company’s advertising expenditures for a

given region in a given month, relative to the total advertising spending for the same region and

month by the company and its main competitor. This share-of-voice advertising measure varies

12

between 0 and 1, and has an average of .74 for free-trial and .79 for regular customers. Table 2

lists summary statistics of the variables in the data set, while Web Appendix A reports the

correlations between the independent variables. We give more details on the control variables

and socio-demographic variables when we discuss the model.

[PLEASE INSERT TABLE 2 ABOUT HERE]

MODEL

We specify a set of equations that incorporates the interrelationships between customers’

usage (flat-rate and pay-per-use) and retention decisions. We account for unobserved customer

heterogeneity, the endogeneity of marketing communication, and selection effects.

Retention

The probability of retention is modeled with a binomial Probit model. Each customer i

decides at the end of every month t after acquisition whether to retain the service (rit = 1) or

disadopt (rit = 0). We write the utility vit of retaining the service as follows:

(1) vit = 0i + 1iTriali + 2iUsageFRit + 3iUsagePPUit + 4iDMit + 5iAdvit +

6iUsageFRit×Triali + 7iUsagePPUit×Triali + 8iDMit×Triali + 9iAdvit×Triali +

10,iInitialit + 11,iFeeitsub + 12,iPenaltyit + 13,iTempit+ 14,i log(Timeit) + it

Thus, the utility of retaining the service at the end of month t is influenced by the dummy Triali

(1 if customer i was acquired with a free-trial; 0 otherwise). This allows us to test whether free-

trial acquisition increases a customer’s churn rate (H1). Other drivers include the customer’s

usage of the flat-rate (UsageFRit, measured in monthly channel zaps divided by 100) and pay-

per-use services (UsagePPUit, measured in number of VODs) in month t. Retention also depends

on the company’s monthly direct-marketing efforts (DMit, the number of direct-marketing

13

contacts received by customer i) and advertising intensity (Advit, the company’s share-of-voice

in customer i's region).4 Equation 1 also includes the interactions between Triali and the usage

and marketing-communication variables to test H2a,b and H3a,b, which posit that the impact of

usage and marketing communication is stronger for free-trial than for regular customers.

Finally, the equation includes a set of control variables. The model accounts for a general

pattern of high defection rates during the first four months of a customer’s tenure through the

dummy variable Initialit. The subscription fee for customer i in month t, Feeitsub, captures the

influence of price on a customer’s retention decision (Ascarza, Lambrecht, and Vilcassim 2012).

Variation in Feeitsub is due to temporary price reductions and the zero-price in the beginning of a

free-trial customer’s tenure. Moreover, we include the variable Penaltyit to account for the fact

that customers were able to cancel their 12-month subscription by paying an early-termination

fee. We assume that customers trade off the termination fee against future subscription fees

within the current contractual period. Hence, Penaltyit equals the termination fee for immediate

disadoption minus the sum of all future subscription fees that the customer would have to pay

during the remaining months of the contractual period. The higher Penaltyit, the more likely it is

that the customer retains the service.

Importantly, the variables Feeitsub and Penaltyit control for the systematically higher

defection rates of free-trial customers, compared to regular customers during the free-trial period.

In this period, free-trial customers face zero subscription and cancelation fees compared to

nonzero subscription and cancelation fees for regular customers. Hence these control variables

allow us to obtain a clean test of H1 via the trial dummy in Equation 1, which equals one for a

free-trial customer even after the free-trial period is over.

We include the monthly average temperature (Temperatureit) to control for seasonality,5

14

and the log of time-since-adoption (Timeit) to accommodate fluctuations in the baseline retention

probability (Prins, Verhoef, and Franses 2009). The model coefficients 0i, …, 14,i are

customer-specific and normally distributed. Finally, the Probit error term it is normally

distributed with a standard deviation set equal to 1 for identification purposes.

Usage of the Flat-Rate Service

We model flat-rate usage (UsageFRit) as a log-log regression to account for the skewed

nature of this variable (Iyengar et al. 2011):

(2) log(UsageFRit+1) = 0i + 1iTriali + 2ilog(UsageFRi,t-1+1) + 3ilog(Feeitsub+1) +

4ilog(Tempit) + 5ilog(Timeit) + it

The free-trial acquisition dummy Triali captures differences in flat-rate usage between free-trial

and regular customers. The lagged dependent variable UsageFRi,t-1 accounts for persistence in

usage behavior. We also include the control variables subscription fee, average temperature, and

time-since-adoption.6 Before taking the logarithm, we add 1 to all variables for which zeros

occur (e.g., Iyengar et al. 2011). 0i, …, 5i are normally distributed customer-specific

coefficients and it is an error term following a normal distribution, N(0, 2).

Usage of the Pay-per-Use Service

We model usage of the pay-per-use component, i.e., a consumer’s monthly number of

VODs, with a zero-inflated Poisson model, in which the zero inflation accommodates the spike

at zero in the VOD usage variable:

(3) UsagePPUit = UsagePPUit* ~ Poisson(λit) with probability qi

UsagePPUit = 0 with probability 1 qi

where qi is the probability that customer i is a potential user of the VOD service, modeled with a

Probit structure with customer-specific normally distributed intercept; UsagePPU*it is the number

15

of VODs watched by customer i in period t, given that the customer is a potential VOD user. The

expected number of VODs, λit, is specified as an exponential function to ensure a positive sign:

(4) λit = exp(γ0i + γ1iTriali + 2iUsagePPUi,t-1 + γ3iFeeitsub + γ4iTempit + γ5ilog(Timeit) +

γ6iCreditit).

Similar to the model for flat-rate usage, the expected number of videos watched is a function of

acquisition mode (Triali), a customer’s past usage (UsagePPUi,t-1), and the variables Feeitsub,

Tempit, and log(Timeit). Creditit is the VOD credit for customer i in month t (measured in €).

This VOD credit is granted by the company for a maximum period of four months to stimulate

service usage. γ0i, …, γ6i are normally distributed customer-specific coefficients.

Customer Heterogeneity

We include customer heterogeneity by modeling all response parameters (intercepts and

slope coefficients) as normally distributed across customers. To incorporate interdependence

between the different model components, we allow for correlations between the intercepts.7 The

expected values of the retention and usage intercepts α0i, 0i, and 0i are functions of the

concomitant customer characteristics age, household size, income (e.g., Rust and Verhoef 2005),

and time-to-adoption (Prins, Verhoef, and Franses 2009; Schweidel, Fader, and Bradlow 2008):

(5) (

E(α0i)

E(β0i)

E(γ0i)) = (

α0,0 + α0,1Agei + α0,2Hhsizei + α0,3Incomei + α0,4Adopttimeiβ0,0 + β0,1Agei + β0,2Hhsizei + β0,3Incomei + β0,4Adopttimeiγ0,0 + γ0,1Agei + γ0,2Hhsizei + γ0,3Incomei + γ0,4Adopttimei

)

where Agei, is the age of customer i (in years, shortly after service launch), Hhsizei, is the size of

customer i’s household (in number of persons), Incomei is the average income in the census

block to which customer i belongs (in €10,000), and Adopttimei is the time-to-adoption of

customer i (measured in months following the launch of the iTV service).

16

Correction for Endogeneity of Marketing Instruments

Endogeneity due to temporal correlation. The first type of endogeneity that we address

involves temporal correlation of DMit and Advit with the error term of the retention equation. For

example, the company may counteract expected increases in the churn rate by boosting its

marketing efforts. Following Park and Gupta (2012) and Schweidel and Knox (2013), we use

Gaussian copulas to model the correlation between marketing and the error term. While classical

methods to correct for endogeneity rely on instrumental variables (IVs) to partial out the

exogenous variation in the endogenous regressors, copulas do not require IVs (Park and Gupta

2012; Schweidel and Knox 2013). In line with Park and Gupta (2012, p. 573), we add the

following regressors to Equation 1:

(6) DMit = Φ−1(HDM(DMit)) and

Advit = Φ−1(HAdv(Advit)),

where -1 is the inverse of the normal cumulative distribution function and HDM(·) and HAdv(·)

represent the empirical cumulative distribution functions of direct marketing and advertising,

respectively.8 For identification purposes, the endogenous regressors must be non-normally

distributed (Park and Gupta 2012), which a Shapiro-Wilk test shows to be the case (direct

marketing: W = .8255, p < .001; advertising: W = .7948, p < .001).

Endogeneity due to cross-sectional correlation. Second, endogeneity may arise from

cross-sectional correlation of the marketing activities with the random intercept. Specifically, the

company may target its direct-marketing efforts on the basis of consumer characteristics

(unobserved to the researcher) that correlate with customers’ churn rates. To address this type of

endogeneity, we follow Mundlak (1978) and include the average number of direct-marketing

contacts per customer, DMi, as a covariate in Equation 1 (e.g., Risselada, Verhoef, and Bijmolt

17

2014). Because the focal firm uses advertising as a mass-communication device, Advit is not

subject to this type of endogeneity. In the estimation, we assess the added value of the Mundlak

correction for cross-sectional endogeneity, on top of the correction for temporal endogeneity.9

Correction for Selection

Selection model. We use two alternative approaches to correct for selection effects: a

selection model and matching. The selection model approach estimates the retention and usage

models jointly with an additional model for consumers’ selection into the free-trial or regular

customer group. By allowing for correlation between the error of the selection equation and the

random intercepts of the usage and retention models, we account for selection effects due to

unobserved variables (Thomas 2001). Since the selection is a single event (sign up as a free-trial

or regular customer), we need to allow for correlations with the random intercepts in retention

and usage rather than with the time-varying error terms. To model whether a customer was

acquired with a free-trial (Triali = 1) or not (Triali = 0), we use a binary probit structure and write

the underlying utility of free-trial acquisition as follows:

(7) wi = ω0 + ω1Agei + ω2Hhsizei + ω3Incomei + ω4Adopttimei + ω5DMi* + ω6Advi

* +

ω7Feei* + ζi

The drivers include age, household size, income, time-to-adoption, and three marketing

variables. DMi* represents the average number of direct-marketing contacts received by customer

i in the three months before signing up. Because the free trial was often promoted in direct-

marketing contacts (e.g., in outbound telephone calls), DMi* likely has a positive impact on

consumers’ awareness of the trial offer. Advi* is the average share-of-voice for customer i in the

three months before signing up. Because advertising usually promoted the regular offer, we

expect Advi* to decrease the probability that a customer was acquired through a free trial. Feei

*

18

refers to the total fees for a regular 12-month subscription at the time of customer i’s signup.

Higher fees for the regular subscription may lead consumers to search longer for a special deal or

push harder when in touch with a customer service agent, such that Feei* should have a positive

effect on the probability of free-trial acquisition. Finally, ζi is a standard-normal error term.

Matching procedure. As an alternative to jointly estimating the selection, retention, and

usage models, we apply a matching procedure which pairs free-trial customers with the most

comparable regular customers. We estimate the models on this matched data set such that

differences in retention and usage behavior can be attributed to the acquisition mode rather than

to differences in sample composition (Gensler, Leeflang, and Skiera 2013; Gensler, Leeflang,

and Skiera 2012). To match subjects, we first estimate Equation 7 to compute customers’

propensities of adopting the service on a free trial. Following Gensler, Leeflang, and Skiera

(2013), we apply a hybrid procedure which combines subjects not only on the basis of their

propensity scores but also on the basis of drivers of the selection process, i.e., the variables DMi*,

Advi*, and Feei

*. Next, we compute the Mahalanobis distances among consumers on the basis of

their propensity scores and these three variables.10 Finally, we use the one-nearest neighbor

algorithm to identify 12,610 pairs of free-trial customers and their most comparable regular

customers (Gensler, Leeflang, and Skiera 2012).

EMPIRICAL RESULTS

We use simulated maximum likelihood with Halton draws to calibrate the usage and

retention models. Before discussing parameter estimates, we check model performance and the

robustness of the results across alternative approaches to tackling endogeneity and selection. We

then evaluate model fit for the selected approach in detail and discuss the estimation results.

19

Robustness Checks and Model Selection

We compare two ways to correct for marketing endogeneity – copula and Mundlak terms

versus a more parsimonious approach with just the copula terms – and two ways to address

selection effects – the selection model versus matching. We discuss model performance and the

robustness of the results across all four combinations. To assess model performance, we evaluate

the retention model’s in-sample fit and out-of-sample fit (cross-sectional and longitudinal) on

three fit measures: hit probability (Gilbride, Allenby, and Brazell 2006), top-decile lift, and Gini

coefficient (Lemmens and Croux 2006). While in- and out-of-sample fit are not suited to

compare models with and without endogeneity correction, they can be used to validate different

approaches that all correct for endogeneity (Ebbes, Papies, and Van Heerde 2011). Table 3

shows that model M1 based on matched samples and including both copula and Mundlak terms

outperforms the other models for seven out of nine fit criteria. Therefore, the rest of our

discussion focuses on model M1. Notice, however, that the four models yield very comparable

outcomes for the hypotheses tests, underlining the robustness of the results.

[PLEASE INSERT TABLES 3-5 ABOUT HERE]

Estimation Results

Table 4 reports in-sample and out-of-sample fit measures for the selected model M1 in

more detail. In addition to the measures for the retention model, we use the correlation between

actual and predicted values as a fit measure for usage of the flat-rate and pay-per-use services.

Overall, the different measures suggest a good in- and out-of-sample model fit, for both free-trial

and regular customers. The fit of the auxiliary propensity model (Equation 7) is adequate (hit

probability: 66.71%) and the parameter estimates, reported in Web Appendix B, are face-valid.

Table 5 presents the parameter estimates for the focal models. For the heterogeneous

20

parameters, we focus on the population means. We first discuss the relationships of the

conceptual framework and then address the effects of the control variables. Hypothesis tests are

one-sided while the remaining tests are two-sided.

Core decision process. The results confirm the expectations for the core retention

decision process. Retention is positively affected by flat-rate usage and usage of the pay-per-use

service (2 = .728, p < .001 and 3 = .039, p < .001, respectively). Usage drives retention

because to the consumer it is an indication of the personal value of the service (e.g., Lemon,

White, and Winer 2002). The impact of both direct marketing (4 = 1.565, p < .001) and

advertising (5 = .310, p < .001) on retention is positive, which is consistent with earlier findings

that marketing communication creates interest in the service and increases retention (e.g., Polo,

Sese, and Verhoef 2011). The significant coefficients of the copula and Mundlak terms suggest

that correcting for marketing endogeneity is indeed required (α15 = –.596, p < .001; α16 = –.017,

p < .001, α17 = –2.111, p < .001).

Differences between free-trial and regular customers. In support of H1, acquisition via

free trials has a direct negative impact on retention (1 = –.278, p < .001), even after controlling

for higher churn during the free-trial period.11 This is in line with the expectation that free-trial

customers are less confident about retaining the service, because they are in a tentative

relationship with the firm and are likely to attribute their subscription to extrinsic incentives (i.e.,

the trial), resulting in lower commitment (Dwyer, Schurr, and Oh 1987).

The positive interaction between the free-trial dummy and usage variables indicates that

the usage effects are stronger for free-trial than for regular customers, which confirms H2a and

H2b (6 = .040, p < .001 for the flat-rate service and 7 = .013, p < .001 for the pay-per-use

service). This is in line with the notion that free-trial customers, because of their lower

21

commitment, are more uncertain about the service’s benefits and therefore rely more on their

usage behavior when making retention decisions (Bolton and Lemon 1999).

In support of H3a and H3b, free-trial customers are more responsive to marketing

communication instruments than regular customers (8 = .133, p < .001 for direct marketing and

9 = .154, p < .001 for advertising). Similar to usage, advertising and direct marketing provide

free-trial customers with cues that make them more secure about the service’s value and assist

them in their retention decision (Mitchell and Olson 1981).12

Free-trial customers use the flat-rate service less intensively than regular customers

(1 = –.269, p < .001). This is in line with the notion that free-trial customers are less committed

to the service’s benefits. Interestingly, however, they use the pay-per-use service more

intensively than regular customers (1 = .113, p < .001). Because free-trial customers make less

use of the flat-rate service, they may dedicate more time to exploring the paid add-on service

(Schary 1971). In addition, they may perceive the free trial as a windfall gain, enticing them to

spend more on the pay-per-use service (e.g., Heilman, Nakamoto, and Rao 2002).

Figure 2 shows plots of the average predicted and observed values for retention (Panel a),

flat-rate usage (Panel b), and usage of the pay-per-use service (Panel c). The plots show that the

model fits the retention and usage data well for both free-trial and regular customers. In Panel a,

the predicted retention rates after the trial look very similar for free-trial and regular customers.

However, the model unveils that these predicted retention rates are obtained in a very different

way for these two customer groups. For free-trial customers, the baseline retention rate is lower,

because the main effect of Triali on retention utility is α1 = –.278. However, this lower baseline

rate is compensated by free-trial customers’ stronger marketing responsiveness (see Table 5) and

by the higher levels of direct marketing they received (see Table 2).

22

[PLEASE INSERT FIGURE 2 ABOUT HERE]

Control variables. The control variables have significant and face-valid effects. TV usage

drops in warmer months for obvious reasons, and the VOD credit increases usage of the VOD

service. In addition, we find positive carry-over effects for both usage components (e.g., Bolton

and Lemon 1999).13 Higher subscription fees reduce a customer’s probability to retain and use

the service (e.g., higher fees reduce customers’ budget and, as a result, decrease their paid VOD

consumption). The concomitant consumer characteristics also play a significant role. For

example, all else equal, larger families are likely to use the service more intensively (both the

flat-rate and pay-per-use component). Finally, the table in Web Appendix C indicates that there

are significant correlations between the random intercepts of the retention and usage equations.

IMPLICATIONS FOR CUSTOMER LIFETIME VALUE AND CUSTOMER EQUITY

To examine how the estimated effects influence CLV, we compare the CLV of free-trial

and regular customers, and compute the elasticities of CLV with respect to changes in marketing

communication efforts and customers’ usage intensities. Throughout the calculations, we use

customer-specific posterior parameter distributions (Train 2009).

For maximum comparability between free-trial and regular customers, we use the same

global means for the marketing variables. Furthermore, to avoid comparing the behavior of

(relatively tentative) free-trial customers in their first three months with (relatively confident)

regular customers, we only analyze those customers who survived the first three months,

retaining 8,624 free-trial customers and 3,145 regular customers. On top of that, in the

simulations, we shock marketing or usage in the first month after the trial period (i.e., in month

4 of a customer’s tenure). We now explain how we derive CLV from customers’ retention and

usage behavior.

23

Calculating Net CLV

We simulate customers’ usage and retention behavior over a three-year time horizon.

Previous research on other high-tech products and services uses the same simulation horizon,

arguing that most of a customer’s value is typically captured during these first three years

(Kumar et al. 2008; Rust, Kumar, and Venkatesan 2011). We compute Net CLV as the total

revenue stemming from a customer’s consecutive retention and usage decisions minus the costs

to acquire and retain that customer. For a given month, revenue consists of the fixed subscription

fee and the pay-per-use fees for watching VODs (corrected for content costs) if the customer

retains the iTV service, or the early-termination fee if the customer disadopts. On the cost side,

we distinguish between direct-marketing and advertising expenditures to acquire and retain the

customer. Other costs are not directly related to the number of customers or their usage intensity

(e.g., the network infrastructure is owned by the company), and can thus be ignored in the

computation of Net CLV. To compute the present value of future cash flows, we use an annual

discount rate of 8.5%, which is common in this industry (e.g., Cusick et al. 2014).14 Web

Appendix D provides the equations for the Net CLV calculations.

Survival times. As a first step towards calculating Net CLV, we investigate customers’

expected survival times during the three-year simulation period. Figure 3 shows that free-trial

customers churn much earlier than regular customers. The survival times are also characterized

by considerable consumer heterogeneity, especially for free-trial customers. Thus, we can expect

substantial differences in Net CLV between but also within the two customer groups.

[PLEASE INSERT FIGURES 3-6 ABOUT HERE]

Net CLV for free-trial and regular customers. The Net CLV results are shown in Figure

4. Panel a presents average revenues, costs, and the resulting CLV for both groups. The major

24

part of revenues comes from subscription fees. The pay-per-use service accounts for 13%

(€36.82) of the total revenues from free-trial customers and 10% (€41.25) of the total revenues

from regular customers. Regular customers, who cannot cancel the service for free during a trial

period, pay a higher expected cancellation fee than free-trial customers (€16.19 versus €7.11).

The costs for acquisition (€170.29 for free-trial and €169.89 for regular customers) and retention

(€10.79 for free-trial and €14.01 for regular customers) are very comparable. After taking into

account all revenues and costs, the resulting Net CLV is on average 59% lower for free-trial

customers than for regular customers (€101.25 versus €244.62).

Panel b shows that there is considerable heterogeneity in Net CLV also within the free-

trial and regular customer groups. Although free-trial customers, on average, have a lower Net

CLV, several of them generate a relatively high value, as becomes clear from the substantial

overlap between the two distributions.

Marketing communication and usage elasticities. The parameter estimates suggest that

free-trial customers are significantly more responsive to changes in marketing communication

and usage levels than regular customers. To quantify whether these differences are also

managerially relevant, we calculate elasticities of Net CLV with respect to marketing

communication and usage. We find these elasticities to be substantially higher for free-trial

customers than for regular customers. As shown in Figure 5, panels a and b, the average

advertising and direct-marketing elasticities of free-trial customers are .26 and .41, respectively.

For regular customers, however, we find average elasticities of only .02 and .08, respectively.15

In a similar vein, we assess the extent to which Net CLV changes in response to an

increase in usage (see Figure 5, panels c and d). For example, the iTV company could enhance

usage by offering access to more channels (flat-rate service) or extending their VOD catalogue

25

(pay-per-use service). We find that the average flat-rate usage elasticity equals .75 for free-trial

customers but only .17 for regular customers. Likewise, the average pay-per-use elasticity is .04

for free-trial customers, yet only .02 for regular customers.

Should a Company offer Free Trials?

One major finding of this research is that the expected value of customers attracted with

free trials is less than the expected value of regular customers (E(CLVfree-trial) = €101.25 versus

E(CLVregular) = €244.62). Consequently, companies may wonder whether offering free trials is

worth the effort. For a free trial to be beneficial, it should attract enough lower-value customers

to at least keep the aggregated CLV, or “customer equity” (CE), constant. But how many

customers does the free-trial need to attract?

To obtain an indication of the necessary number of additional customers, we can use the

results of the CLV calculations. According to the estimates, the 8,624 free-trial customers who

survive the free-trial period generate a total expected CE of €873,180 (8,624 E(CLVfree-trial)).

To generate the same equity without the free trial, the firm would have had to attract only 3,570

regular customers because 3,570 E(CLVregular) = 8,624 E(CLVfree-trial) = €873,180. Hence,

offering consumers a free trial rather than a regular contract should lead to = 8,624/3,570 =

E(CLVregular)/E(CLVfree-trial ) = 2.42 times more customers to keep CE constant. If a company

believes a free trial is able to attract more new customers than this threshold, we recommend

offering the free trial rather than the regular contract because the expected expansion of the

customer base will compensate for the customers’ lower CLVs; otherwise the company should

not offer the trial.

Importantly, is not a fixed number but depends, to some extent, on the company’s

marketing efforts. Specifically, the company can lower the threshold by increasing its marketing

26

efforts after acquisition, making it easier for the free trial to break even in terms of CE. Indeed,

the results indicate that higher marketing efforts lead to a smaller difference between E(CLVfree-

trial) and E(CLVregular), such that the threshold = E(CLVregular)/ E(CLVfree-trial) decreases. We

illustrate this principle in Figure 6. For example, for direct-marketing (versus advertising) efforts

20% above the observed average, drops from 2.42 to 2.26 (versus 2.36). Thus, with increased

direct-marketing (advertising) expenditures, it suffices if the trial is only 2.26 (2.36) more

successful at attracting customers than the regular offer. These insights help companies to trade

off acquisition and retention efforts (e.g., Reinartz, Thomas, and Kumar 2005). For example, a

company may decide to boost retention expenditures when a trial’s acquisition targets are not

met. Vice versa, a well-designed free-trial promotion may lead to a higher-than-expected

increase in customers, which may justify lower retention efforts afterwards.

[PLEASE INSERT FIGURE 6 AND TABLE 6 ABOUT HERE]

Improving Retention and Customer Equity after Acquisition

When a company has acquired customers with free trials and regular subscriptions, a key

managerial question is how to improve consumers’ retention behavior and customer equity.

Because free-trial customers are worth less than regular customers, some managers may decide

to invest less in free-trial customers. However, their higher responsiveness to marketing

communication actually suggests otherwise. Therefore, we recommend managers to pay specific

attention to free-trial customers. To illustrate this recommendation, we simulate the impact on

retention and CE of alternative targeting strategies after customer acquisition (see Table 6).

Suppose a manager has an additional monthly direct-marketing budget that enables her to

increase direct-marketing efforts by €0.5 per customer for 20% of the customers. Given the

company’s total customer base of 160,000 customers at the end of the observation period, this

27

corresponds to a substantial marketing investment of about €576,000 (160,000 × 20% × €0.5 ×

36 months). The question then becomes how to select the 20% target customers. Using the

customers in the simulation data set, we consider different scenarios. In scenario 1, the manager

simply selects customers from the company’s database by chance, yielding an increase in

retention (158.0%) and Net CE (115.9%) for the 20% targeted customers. The impact across the

entire customer base is also substantial: 31.6% for retention, 23.0% for Net CE.

Alternatively, bearing in mind our finding that acquisition mode influences customers’

response to marketing communication, a manager could follow scenario 2 and target a random

subset of free-trial customers. Such a strategy would improve retention and Net CE of the

selected customers even more (186.7% for retention and 136.0% for Net CE). These returns

stand in sharp contrast to scenario 3 in which the manager invests only in a random selection of

regular customers, leading to an increase of 95.6% for retention and 75.6% for Net CE. Finally,

a manager could opt for scenario 4 in which individual-level response coefficients are used to

select customers with the highest responsiveness to direct-marketing. Such an optimal strategy

would improve retention and Net CLV by 181.4% and 132.9% for targeted customers, and

36.3% and 26.6% for all customers. Notice that targeting free-trial customers (scenario 2) leads

to results that come close to the optimal ones.

DISCUSSION

Recent research suggests that the way in which customers are acquired may have an

enduring impact on their behavior, even long after adoption (e.g., Villanueva, Yoo, and Hanssens

2008). While some studies have found differences between customers attracted with sales

promotions and regular customers (Anderson and Simester 2004; Lewis 2006), the challenge of

retaining customers acquired through free-trial promotions is not well understood. Crucial

28

questions have remained unanswered: How do free-trial customers differ from regular customers

in their retention rates, marketing responsiveness, and lifetime value? Should free-trial customers

be managed in a different way?

To answer these questions, we model a customer’s decisions to use and retain the service

– decisions that influence CLV – and examine how free-trial acquisition affects this decision

process. We explicitly account for selection effects (i.e., customers attracted with free trials may

differ intrinsically from other customers) and for the endogeneity of marketing communication.

The model is estimated on a unique panel data set, covering monthly usage (of flat-rate and

pay-per-use services) and retention decisions for 16,512 customers of an interactive digital TV

service provider. We then use the parameter estimates to simulate CLV and quantify its

sensitivity to changes in marketing communication and usage rates.

Throughout the analyses, we find strong evidence for systematic differences in behavior

between free-trial and regular customers. In line with buyer-seller relationship theory, which

suggests that free trials may slow down the relationship formation process, free-trial promotions

are associated with higher defection rates, even beyond the free-trial period. Similarly, flat-rate

usage – an important driver of customers’ retention decisions – turns out to be lower among free-

trial subscribers. Interestingly, however, free-trial customers have a higher pay-per-use

consumption rate. Apparently, they readily reallocate the time they gain (due to their lower flat-

rate consumption) and the money they save (due to the free trial) to the pay-per-use service.

These results add to the growing understanding of the role of usage in the value generation

process (e.g., Bolton and Lemon 1999; Iyengar et al. 2011).

Managerial Implications

Our study has a number of key managerial implications. Because of their higher churn

29

rate, free-trial customers are worth considerably less than regular customers. Specifically, we

find the Net CLV of free-trial customers to be 59% lower than that of regular customers.

Managers and business analysts may thus have to temper profit expectations if the customer base

includes a substantial share of free-trial subscribers. At the same time, we find free-trial

customers to be more “malleable” than regular customers. Because they have a less-developed

relationship with the firm, free-trial customers are more uncertain about the service benefits. As a

result, they rely more on marketing communication and their own usage behavior when deciding

whether or not to retain the service. We find that, compared to regular customers, the Net CLV

of free-trial customers is much more responsive to direct marketing, advertising, flat-rate usage,

and pay-per-use usage.

Therefore, companies should target direct marketing and advertising more to free-trial

than to regular customers. In these marketing communications, it is advised to remind free-trial

customers about their usage rates, especially when they are high, to make these customers even

more likely to retain the service. For example, mobile phone providers could use direct

marketing to communicate for how many minutes the customer has used the flat-rate service in

the past period, cementing the relevance of the service. Companies could also try to enhance the

actual usage of the service. For example, in the context of digital TV, firms could enhance usage

opportunities by offering recorded TV shows or providing apps to watch TV on mobile devices.

Other service providers can estimate our retention and usage models to calculate the

expected CLV for every customer, acquired by a free trial or not. They can then focus their

marketing efforts on those customers with the highest expected return on marketing investment.

If the estimation of the models is not feasible, our results suggest managers can rely on usage

intensity as an indicator of customer value and a criterion for marketing allocation decisions.

30

This allows managers to distinguish between (likely) high- and low-value customers, especially

within the free-trial group, because this group is most responsive to usage levels.

Our analysis also offers insights on the factors playing a role when considering the launch

of a free-trial campaign. We show how the company can obtain an indication of the free trial’s

required expansion effect on the customer base to compensate for free-trialists’ lower lifetime

value and make the free-trial campaign worthwhile. Specifically, relative to a situation with only

the regular offer, the number of adopters should increase by a factor that can be calculated as the

expected CLV of regular customers divided by the expected CLV of free-trial customers. If the

company expects the free trial to attract more customers than this factor prescribes, the trial offer

may be a viable strategy. Importantly, the company can lower this break-even factor through

targeted retention efforts after acquisition.

Future Research

Our work is the first to investigate the effects of free-trial acquisition on customer

behavior and CLV, and therefore leaves several opportunities for further research. First, because

this study is based on observational data, we need to control for selection effects – a challenge

intrinsic to many studies examining the behavior of distinct customer groups (e.g., Gensler,

Leeflang, and Skiera 2012). Alternatively, researchers may use controlled field experiments. One

setting particularly suited for field experiments are online services, where individual consumers

can be randomly assigned to different offers, and where the likelihood of consumers becoming

aware of alternative offers is limited. At the same time, such an approach would still not

guarantee full comparability of free-trial and regular customers because the researcher cannot

force consumers to accept an offer.

Second, controlled experiments would also allow for the uncovering of the underlying

31

psychological processes that are likely to play a role. One interesting observation worthy of

further attention is that free-trial customers are more heterogeneous than regular customers, for

example in terms of CLV and responsiveness to marketing and usage. We speculate that the

segment of free-trial customers includes not only people who are truly doubtful about their

commitment, but also opportunistic consumers who would have adopted the service anyhow and

just subscribe to the trial to enjoy the free months. Future research can further investigate the

heterogeneity in the motivations of free-trial and regular consumers.

Third, future work could replicate this research in other contexts. Many Internet

companies currently operate under a freemium model, in which customers can choose between a

“free service” and a “premium service” which comes at a fee but provides enhanced functionality

(Pauwels and Weiss 2008). For example, the on-demand streaming service Spotify offers

temporary one-month free trials for their premium service, next to its (permanent) free service.

Fourth, future research may give a more complete account of the profitability

implications of free-trial and regular acquisition. For example, for a news site, usage may be a

source of advertising revenues (e.g., the number of page views may determine advertising

income). Furthermore, it may be worthwhile to also include other value-creating behaviors

(Verhoef, Reinartz, and Krafft 2010). Notably, one could examine whether free-trial and regular

customers differ in the extent to which they engage in word-of-mouth communication.

In summary, we uncover the implications of free-trial acquisition for customer retention

and CLV. We show how service usage and marketing activities drive consumers’ retention

decisions and CLV, and how free-trial acquisition moderates these relationships. Our findings

offer managers new insights on how to retain customers acquired through free trials.

32

REFERENCES

Anderson, Eric. T. and Duncan I. Simester (2004), “Long-Run Effects of Promotion Depth on

New Versus Established Customers: Three Field Studies,” Marketing Science, 23 (1), 4-

20.

Ascarza, Eva, Anja Lambrecht and Naufel Vilcassim (2012), “When Talk Is “Free”: The Effect

of Tariff Structure on Usage Under Two- and Three-Part Tariffs,” Journal of Marketing

Research, 49 (6), 882-99.

Bawa, Kapil and Robert Shoemaker (2004), “The Effects of Free Sample Promotions on

Incremental Brand Sales,” Marketing Science, 23 (3), 345-63.

Blattberg, Robert C., Edward C. Malthouse and Scott A. Neslin (2009), “Customer Lifetime

Value: Empirical Generalizations and Some Conceptual Questions,” Journal of

Interactive Marketing, 23 (2), 157-68.

Bolton, Ruth N. and Katherine N. Lemon (1999), “A Dynamic Model of Customers’ Usage of

Services: Usage as an Antecedent and Consequence of Satisfaction,” Journal of

Marketing Research, 36 (2), 171-86.

Brucks, Merrie (1985), “The Effects of Product Class Knowledge on Information Search

Behavior,” Journal of Consumer Research, 12 (1), 1-16.

Chan, Tat Y., Chunhua Wu and Ying Xie (2011), “Measuring the Lifetime Value of Customers

Acquired from Google Search Advertising,” Marketing Science, 30 (5), 837-50.

Cusick, Philip, Richard Choe, Eric Pan, and Ava Zhang (2014), “DirecTV: Takeways from DTV

Management Meetings - US and Latam [J.P. Morgan Research],” Thomson ONE Banker

Analytics (accessed 30 September 2014), http://banker.thomsonib.com/

Dodson, Joe A., Alice M. Tybout and Brian Sternthal (1978), “Impact of Deals and Deal

33

Retraction on Brand Switching,” Journal of Marketing Research, 15 (1), 72-81.

Dwyer, F. Robert, Paul H. Schurr and Sejo Oh (1987), “Developing Buyer-Seller Relationships,”

Journal of Marketing, 51 (2), 11-27.

Ebbes, Peter, Dominik Papies and Harald J. Van Heerde (2011), “The Sense and Non-Sense of

Holdout Sample Validation in the Presence of Endogeneity,” Marketing Science, 30 (6),

1115-22.

Gedenk, Karen and Scott A. Neslin (1999), “The Role of Retail Promotion in Determining

Future Brand Loyalty: Its Effect on Purchase Event Feedback,” Journal of Retailing, 75

(4), 433-59.

Gensler, Sonja, Peter Leeflang and Bernd Skiera (2012), “Impact of Online Channel Use on

Customer Revenues and Costs to Serve: Considering Product Portfolios and Self-

Selection,” International Journal of Research in Marketing, 29 (2), 192-201.

_____ (2013), “Comparing Methods to Separate Treatment from Self-Selection Effects in an

Online Banking Setting,” Journal of Business Research, 66 (9), 1272-78.

Gilbert, Daniel T., Timothy D. Wilson, Elizabeth C. Pinel, Stephen J. Blumberg, and Thalia P.

Wheatley (1998), “Immune Neglect: A Source of Durability Bias in Affective

Forecasting,” Journal of Personality and Social Psychology, 75 (3), 617-38.

Gilbert, Daniel. T. and Jane E. J. Ebert (2002), “Decisions and Revisions: The Affective

Forecasting of Changeable Outcomes,” Journal of Personality and Social Psychology, 82

(4), 503-14.

Gilbride, Timothy J., Greg M. Allenby and Jeff D. Brazell (2006), “Models for Heterogeneous

Variable Selection,” Journal of Marketing Research, 43 (3), 420-30.

Heilman, Carrie M., Kent Nakamoto and Ambar G. Rao (2002), “Pleasant Surprises: Consumer

34

Response to Unexpected In-store Coupons,” Journal of Marketing Research, 39 (2), 242-

52.

Heiman, Amir, Bruce McWilliams, Zhihua Shen, and David Zilberman (2001), “Learning and

Forgetting: Modeling Optimal Product Sampling over Time,” Management Science, 47

(4), 532-46.

Hoch, Stephen J. and John Deighton (1989), “Managing What Consumers Learn from

Experience,” Journal of Marketing, 53 (2), 1-20.

Iyengar, Raghuram, Kamel Jedidi, Skander Essegaier, and Peter J. Danaher (2011), “The Impact

of Tariff Structure on Customer Retention, Usage, and Profitability of Access Services,”

Marketing Science, 30 (5), 820-36.

Jain, Dipak, Vijay Mahajan and Eitan Muller (1995), “An Approach for Determining Optimal

Product Sampling for the Diffusion of a New Product,” Journal of Product Innovation

Management, 12 (2), 124-35.

Johnson, Michael D. and Fred Selnes (2004), “Customer Portfolio Management: Toward a

Dynamic Theory of Exchange Relationships,” Journal of Marketing, 68 (2), 1-17.

Kumar, V., Rajkumar Venkatesan, Tim Bohling, and Denise Beckmann (2008), “Practice Prize

Report—The Power of CLV: Managing Customer Lifetime Value at IBM,” Marketing

Science, 27 (4), 585-99.

Lemmens, Aurélie and Christophe Croux (2006), “Bagging and Boosting Classification Trees to

Predict Churn,” Journal of Marketing Research, 43 (2), 276-86.

Lemon, Katherine N., Tiffany Barnett White and Russell S. Winer (2002), “Dynamic Customer

Relationship Management: Incorporating Future Considerations into the Service

Retention Decision,” Journal of Marketing, 66 (1), 1-14.

35

Lewis, Michael (2006), “Customer Acquisition Promotions and Customer Asset Value,” Journal

of Marketing Research, 43 (2), 195-203.

Luo, Xueming, Sascha Raithel and Michael A Wiles (2013), “The Impact of Brand Rating

Dispersion on Firm Value,” Journal of Marketing Research, 50 (3), 399-415.

Mitchell, Andrew A. and Jerry C. Olson (1981), “Are Product Attribute Beliefs the Only

Mediator of Advertising Effects on Brand Attitude?,” Journal of Marketing Research, 18

(3), 318-32.

Mundlak, Yair (1978), “On the Pooling of Time Series and Cross Section Data,” Econometrica,

46 (1), 69-85.

Palmatier, Robert W., Mark B. Houston, Rajiv P. Dant, and Dhruv Grewal (2013), “Relationship

Velocity: Toward a Theory of Relationship Dynamics,” Journal of Marketing, 77 (1), 13-

30.

Park, Sungho and Sachin Gupta (2012), “Handling Endogenous Regressors by Joint Estimation

Using Copulas,” Marketing Science, 31 (4), 567-86.

Pauwels, Koen and Allen Weiss (2008), “Moving from Free to Fee: How Online Firms Market

to Change Their Business Model Successfully,” Journal of Marketing, 72 (3), 14-31.

Polo, Yolanda, F. Javier Sese and Peter C. Verhoef (2011), “The Effect of Pricing and

Advertising on Customer Retention in a Liberalizing Market,” Journal of Interactive

Marketing, 25 (4), 201-14.

Prins, Remco, Peter C. Verhoef and Philip Hans Franses (2009), “The Impact of Adoption

Timing on New Service Usage and Early Disadoption,” International Journal of

Research in Marketing, 26 (4), 304-13.

Reinartz, Werner J., J. S. Thomas and V. Kumar (2005), “Balancing Acquisition and Retention

36

Resources to Maximize Customer Profitability,” Journal of Marketing, 69 (1), 63-79.

Risselada, Hans, Peter C. Verhoef and Tammo H.A. Bijmolt (2014), “Dynamic Effects of Social

Influence and Direct Marketing on the Adoption of High-Technology Products,” Journal

of Marketing, 78 (2), 52-68.

Rust, Roland T., V. Kumar and Rajkumar Venkatesan (2011), “Will the Frog Change into a

Prince? Predicting Future Customer Profitability,” International Journal of Research in

Marketing, 28 (4), 281-94.

Rust, Roland T. and Peter C. Verhoef (2005), “Optimizing the Marketing Interventions Mix in

Intermediate-Term CRM,” Marketing Science, 24 (3), 477-89.

Schary, Philip B. (1971), “Consumption and the Problem of Time,” Journal of Marketing, 35 (2),

50-55.

Schmitt, Philipp, Bernd Skiera and Christophe Van den Bulte (2011), “Referral Programs and

Customer Value,” Journal of Marketing, 75 (1), 46-59.

Schweidel, David A., Peter S. Fader and Eric T. Bradlow (2008), “Understanding Service

Retention Within and Across Cohorts Using Limited Information,” Journal of Marketing,

72 (1), 82-94.

Schweidel, David A. and George Knox (2013), “Incorporating Direct Marketing Activity into

Latent Attrition Models,” Marketing Science, 32 (3), 471-87.

Scott, Carol A. (1976), “The Effects of Trial and Incentives on Repeat Purchase Behavior,”

Journal of Marketing Research, 13 (3), 263-69.

Steffes, Erin M., B.P.S. Murthi and Ram C. Rao (2011), “Why are some Modes of Acquisition

more Profitable: A Study of the Credit Card Industry,” Journal of Financial Services

Marketing, 16 (2), 90-100.

37

Thomas, Jacquelyn S. (2001), “A Methodology for Linking Customer Acquisition to Customer

Retention,” Journal of Marketing Research, 38 (2), 262-68.

Train, Kenneth (2009), Discrete Choice Methods with Simulation: Cambridge University Press.

Van Doorn, Jenny, Katherine N. Lemon, Vikas Mittal, Stephan Nass, Doreen Pick, Peter Pirner,

and Peter C. Verhoef (2010), “Customer Engagement Behavior: Theoretical Foundations

and Research Directions,” Journal of Service Research, 13 (3), 253-66.

Verhoef, Peter C. and Bas Donkers (2005), “The Effect of Acquisition Channels on Customer

Loyalty and Cross-Buying,” Journal of Interactive Marketing, 19 (2), 31-43.

Verhoef, Peter C., Werner J. Reinartz and Manfred Krafft (2010), “Customer Engagement as a

New Perspective in Customer Management,” Journal of Service Research, 13 (3), 247-

52.

Villanueva, Julian, Shijin Yoo and Dominique M. Hanssens (2008), “The Impact of Marketing-

Induced Versus Word-of-Mouth Customer Acquisition on Customer Equity Growth,”

Journal of Marketing Research, 45 (1), 48-59.

38

FOOTNOTES

1 According to economic theory, the lower marginal consumption cost of a two-part (as opposed to pay-per-use) tariff

structure should lead to higher usage rates. Iyengar et al. (2011) explain their counterintuitive finding by pointing out that

partitioned prices draw consumer’s attention and make them more price-sensitive. 2 The remaining 29% is captured by smaller players, operating through satellite or via cable. 3 We measure a customer’s active use of the flat-rate service by using monthly zaps. We also have a partially observed

measure of a customer’s passive use. Specifically, for a period of just 6 months, we observe the variable hours, capturing

the monthly number of hours that the set-top box was switched on. However, this may not be an accurate indication of the

time that the customer actively watched TV. In fact, it was technically possible to switch off the TV while leaving the set-

top box switched on, making this variable less than ideal. For the months in which we have both zaps and hours, we find a

significant positive correlation of .66 (p < .01). 4 We also estimate a retention model with lagged effects of DM and Adv., leaving model fit virtually unaffected (Hit

probability = .0002, Top-decile lift = .0418, Gini = .0016). We thus opt for the more parsimonious model (1). 5 To account for differences in the length of each month and the occurrence of holidays, we also estimate retention and

usage models with monthly dummy variables. However, the results are very similar and model fit hardly changes (Hit

probability = .0004, Top-decile lift = .0748, Gini = .0009, Flat-rate = .0242, Pay-per-use = .0031). In addition, the

model becomes cumbersome to estimate with 66 seasonality parameters (11 heterogeneous month coefficients in three

equations). We therefore stick to the more parsimonious model (1). 6 We also estimate models in which we allow direct marketing and advertising to affect service usage but this does not

lead to a notable improvement in model performance (Flat-rate = .0020, Pay-per-use = .0137). 7 To check whether we should structurally account for any correlation between the slope coefficients, we inspect the

correlations between the consumer-specific posterior slopes (Train 2009) and find the mean absolute correlation to be very

small (.0084). Further, the low correlation between the residuals of the usage models (.0970) suggests that, after

controlling for cross-sectional correlation, there is not much interdependency between the error terms left. 8 Following Park and Gupta (2012, footnote 3 and expression 10), we use empirical instead of estimated densities to

generate DMit and Advit. To keep estimations tractable, we treat DM as a continuous variable and use regular standard

errors instead of bootstrapped ones (shown to be virtually the same, see Park and Gupta 2012). We test the retention model

on a simulated data set with endogenous regressors and recover the true parameters well. 9 In the models in which we correct for both temporal and cross-sectional correlation, we compute DMit using the mean-

centered direct-marketing variable: mean-centering ensures that we only retain that part of the original direct-marketing

variable that may be correlated with the error term it in Equation 1 (Mundlak 1978). 10 From the 12,612 free-trial customers in the data set, we eliminate two customers because their propensity scores do not

lie within the region of common support, a region defined as the overlap between the propensity score distributions of the

two customer groups (Gensler, Leeflang, and Skiera 2012). 11 Because free-trial customers may be particularly likely to churn in their first three months, we re-estimate the model

only with customers who “survived” the first three months. The results are robust to this alternative specification

(1 = .040, z = 3.390, p < .001). 12 We test whether the marketing response estimates are affected by the inclusion of customers who churn already during

the first months after acquisition. Leaving out these customers yields comparable results (8 = .046, z = 2.30, p < .05;

9 = .08, z = 5.11, p < .001, both one-tailed). 13 The carry-over coefficients for the two usage components are not directly comparable because we use a log-log

regression for flat-rate usage and a zero-inflated Poisson model for the pay-per-use service. Nonetheless, the relative

magnitude of the coefficients suggests that the flat-rate service is characterized by higher state-dependence than the pay-

per-use service. This is consistent with the idea that consumers get used to watching the same television shows or series

(flat-rate usage) but rather irregularly consume unrelated content from the VOD service (pay-per-use). 14 The relative results remain the same when we vary the simulation horizon or discount rate. 15 We compute customer-specific arc elasticities for a 1% increase in the focal variable in the first period after acquiring

the customer, and apply a 95% Winsorization on estimated Net CLV elasticities to make sure that the summary statistics

are not driven by outliers (Luo, Raithel, and Wiles 2013). Differences in elasticities between free-trial customers and

regular customers are higher than what perhaps could be expected on the basis of the estimated coefficients (e.g., .310 for

regular customers, and .310 + .154 for free-trial customers in the case of advertising). However, due to differences in the

Net CLV base levels between the two customers groups, the same absolute lift in Net CLV represents a much higher

percentage lift for a free-trial customer than for a regular customer.

39

Table 1

COMPARISON OF THIS PAPER WITH EXISTING STUDIES ON THE RELATIONSHIP BETWEEN ACQUISITION MODE AND CUSTOMER BEHAVIOR

Acquisition Mode

Inclusion

Usage

Behavior

Impact of Acquisition

Mode on Response to

Retention Efforts

Correction

for Selection

Relevant Findings with Regard to Impact on Customer

Behavior

Channel Impact

Verhoef and

Donkers (2005)

Acquisition via Internet, direct-

response advertising, and direct

marketing (and more)

- - - Acquisition through direct mail or direct-response commercials

leads to lower retention rates than online acquisition.

Steffes, Murthi, and

Rao (2008)

Online and direct-marketing

channels (direct mail, direct

selling, telesales)

- - - Acquisition through direct mail yields highest retention rates.

Steffes, Murthi,

and Rao (2011)

Online vs. direct-marketing

channels (direct mail, direct

selling, telesales)

- - Online channel leads to customers with higher transaction

amounts.

Referral Impact

Villanueva, Yoo,

and Hanssens

(2008)

Word-of-mouth vs. marketing-

induced acquisition

- - - Customers acquired through word-of-mouth stay longer with the

firm and generate more future referrals than other customers.

Chan, Wu, and Xie

(2011)

Acquisition from Google

advertising

- - - Customers referred from Google search advertising have greater

retention and transaction rates than other customers.

Schmitt, Skiera,

and Van den Bulte

(2011)

Acquisition via referral

program

- - - Customers attracted via referral program have higher retention

rates than other customers.

Pricing/Sales

Promotion Impact

Anderson and

Simester (2004)

Acquisition with price discount - - - Promotionally acquired customers choose cheaper products but

buy more (often).

Lewis (2006) Acquisition with price discount - - - Promotionally acquired customers have lower retention rates.

Iyengar et al.

(2011)

Acquisition under two-part vs.

pay-per-use pricing

- Not necessarya Customers acquired under two-part pricing have lower usage

and retention rates than customers charged on pay-per-use basis.

This paper Acquisition with free trial Free-trial customers have higher churn rates and lower CLV, but

are more responsive to marketing communication and to own

usage behavior.

a Customers are randomly assigned to one of two pricing structures.

40

Table 2

DESCRIPTIVE STATISTICS FOR FREE-TRIAL AND REGULAR CUSTOMERS

Free-Trial Customers Regular customers

Mean (SD) Mean (SD)

Dependent variables

Retention probability .93 (.25) .96 (.19)

Usage of flat-rate service (channel zaps) 169.35 (176.90) 189.28 (183.80)

Usage of pay-per-use service (number of videos-on-demand) .73 (2.19) .58 (1.91)

Marketing communication

Direct marketing (number of direct-marketing contacts) .34 (.55) .16 (.41)

Advertising (share-of-voice) .74 (.28) .79 (.26)

Customer-specific variables

Age (years) 46.22 (12.64) 44.89 (12.60)

Household size 2.98 (1.48) 2.88 (1.53)

Annual income (in €10,000) 2.44 (.56) 2.43 (.59)

Time-to-adoption (in months following the launch of the service) 13.14 (2.45) 12.38 (2.20)

Control variables

Monthly subscription fee (in €) 9.66 (7.80) 13.22 (4.85)

Cancellation penalty (Termination Fees – Future Fees, in €) -59.80 (68.08) -8.14 (25.02)

VOD credit (in €) 5.50 (8.90) 1.38 (5.08)

Temperature (in °C) 12.82 (4.97) 12.83 (4.99)

Time-since-adoption (in months) 5.34 (3.52) 5.95 (3.68)

Direct-marketing contacts prior to acquisition .62 (.61) .26 (.48)

Advertising (share-of-voice) prior to acquisition .73 (.27) .76 (.27)

Total fees for regular 12-month contract at the time of acquisition (in €) 193.83 (50.04) 181.97 (44.79)

Number of customers 12,612 3,900

41

Table 3

SELECTION OF MODEL FOR RETENTION DECISION

M1

(selected model)

M2 M3 M4

Model specification

Correction for selection Matching Matching Joint estimation Joint estimation

Correction for marketing endogeneity Copula + Mundlak Copula only Copula + Mundlak Copula only

Hypothesis testing

H1: Trial effect on Retention < 0 .278 *** .106 *** .255 *** .066 ***

H2a: UsageFR × Trial effect > 0 .040 *** .107 *** .008 .056 ***

H2b: UsagePPU × Trial effect > 0 .013 ** .015 ** .008 * .013 **

H3a: DM × Trial effect > 0 .133 *** .057 *** .096 *** .051 ***

H3b: Adv × Trial effect > 0 .154 *** .150 *** .150 *** .134 ***

Model fit

Hit probabilitya

In-sample .917 .914 .908 .904

Out-of-sample (cross-sectional) .922 .913 .904 .904

Out-of-sample (longitudinal) .953 .918 .948 .952

Top-decile lift

In-sample 5.853 5.308 5.776 5.153

Out-of-sample (cross-sectional) 7.111 5.130 5.800 5.306

Out-of-sample (longitudinal) 4.244 3.055 4.453 5.219

Gini coefficient

In-sample .755 .713 .748 .707

Out-of-sample (cross-sectional) .826 .711 .752 .706

Out-of-sample (longitudinal) .578 .402 .568 .664

Notes: *** p < .001; ** p < .05; * p < .1 (all one-sided). a We highlight the best model fit in bold. To compute out-of-sample fit measures, we reestimate the model on a random cross-

section of 70% of the customers, or on the first 70% of the longitudinal observations, and compute model fit for the holdout

sample.

42

Table 4

MODEL FIT FOR SELECTED MODEL (M1)

Retention Flat-rate usage Pay-per-use

Hit probability Top-decile lift Gini

coefficient

Correlation between

actual and predicted

Correlation between

actual and predicted

In-sample fit

Free-trial customers .899 5.696 0.750 .811 .750

Regular customers .932 5.634 0.750 .815 .740

Out-of-sample fit (cross-sectional)a

Free-trial customers .906 6.946 0.821 .809 .724

Regular customers .936 7.052 0.819 .815 .727

Out-of-sample fit (longitudinal)b

Free-trial customers .948 4.526 0.599 .751 .673

Regular customers .958 3.785 0.545 .778 .593

a We reestimate the model on a random draw of 70% of the customers and compute model fit for the holdout sample. b We reestimate the model on the first 70% of the longitudinal observations, and predict usage and retention for the holdout sample.

43

Table 5

ESTIMATION RESULTSa

Population Mean

Standard Deviation

Estimate (SE)

Estimate (SE)

Retention

Intercept 3.034 (.006) .025 (.018)

Age .002 (.000)

Household size .037 (.002)

Income .059 (.002)

Time-to-adoption .002 (.000)

Trial H1: effect < 0 .278 (.008) .016 (.021)

Flat-rate usage .728 (.006) .333 (.006)

× Trial H2a: effect > 0 .040 (.009) .159 (.015) Pay-per-use usage .039 (.004) .015 (.005)

× Trial H2b: effect > 0 .013 (.005) .010 (.007)

Direct marketing 1.565 (.012) .190 (.034) × Trial H3a: effect > 0 .133 (.014) .086 (.043)

Advertising .310 (.008) .000 (.027)

× Trial H3b: effect > 0 .154 (.011) .080 (.033)

Copula correction term for DM .596 (.005)

Copula correction term for Adv .017 (.002)

Mundlak correction term for DM 2.111 (.014)

Control variables

Initial period 1.377 (.007) .040 (.016)

Monthly subscription fee .008 (.000) .001 (.001)

Cancellation penalty .003 (.000) .000 (.000)

Temperature .012 (.000) .000 (.001)

log(Time-since-adoption) .572 (.004) .022 (.014)

Flat-rate usage (log zaps)

Intercept 5.609 (.006) .881 (.005) Age .000 (.000)

Household size .036 (.002)

Income .034 (.002)

Time-to-adoption .048 (.000)

Trial .269 (.008) .081 (.014)

Control variables

log(Monthly subscription fee) .033 (.002) .004 (.004)

log(Temperature) .417 (.002) .016 (.004)

log(Time-since-adoption) .319 (.003) .340 (.004)

lagged log(Flat-rate usage) .242 (.001) .005 (.002)

log(sigma) .007 (.002)

Pay-per-use (VODs)

Intercept (incidence) -.140 (.006) 1.410 (.004)

Intercept (amount) .243 (.007) 1.245 (.018)

Age .014 (.000)

Household size .090 (.002)

Income .042 (.003)

Time-to-adoption .018 (.001)

Trial .113 (.010) .016 (.015)

Control variables

Monthly subscription fee .006 (.000) .001 (.000)

VOD credit .005 (.001) .008 (.001)

Temperature .005 (.000) .015 (.001)

log(Time-since-adoption) .053 (.003) .156 (.004)

lagged Pay-per-use .023 (.000) .008 (.000)

Log-likelihood: 310,958; N: 196,251; BIC: 622,892; AIC: 622,077. a Numbers in bold are significant at the p < .05 level (two-sided).

44

Table 6

IMPROVING RETENTION RATES AND CUSTOMER EQUITY

Impact for

targeted customers (20%)

Impact for

entire customer base (100%)

Impact on

retentiona

Impact on

Net CEb

Impact on

retentiona

Impact on

Net CEb

Increase direct marketing efforts by €0.5

for 20% of the customers

Scenario 1: No targeting (random) 158.0% 115.9% 31.6% 23.0%

Scenario 2: Target free-trial

customers (random)

186.7% 136.0% 37.3% 27.2%

Scenario 3: Target regular

customers (random)

95.6% 75.6% 19.1% 15.1%

Scenario 4: Target customers based on highest individual-

level response coefficients

181.4% 132.9% 36.3% 26.6%

a Number of retained customers in month 36 after the start of the free-trial program. b Net Customer Equity for retained customers in month 36 after the start of the free-trial program.

45

Figure 1

CONCEPTUAL FRAMEWORK

Retention

Usage of

pay-per-use service

Usage of

flat-rate service

CLV

H1

H3a

H3b

H2b

H2a

Direct marketing Advertising

Acquisition mode:

Free-trial or regular

46

Figure 2

MODEL FIT

47

Figure 3

DISTRIBUTION OF EXPECTED SURVIVAL TIMES

FOR FREE-TRIAL AND REGULAR CUSTOMERS

Notes: To calculate survival times, we consider a maximum time horizon of

three years (36 months).

48

Figure 4

NET CUSTOMER LIFETIME VALUE FOR FREE-TRIAL AND REGULAR CUSTOMERS

49

Figure 5

RESPONSIVENESS TO MARKETING COMMUNICATION

Notes: We compute arc elasticities for a 1% increase in, respectively, advertising, direct marketing, and usage levels in month 4 after acquiring

the customer (i.e., after the trial period).

50

Figure 6

REQUIRED IMPACT OF FREE TRIALS ON CUSTOMER BASE EXPANSION

FOR DIFFERENT LEVELS OF MARKETING

Notes: represents the threshold value to keep customer equity constant, and can be interpreted as how much more successful the free trial should be at attracting new customers compared to the regular offer.

51

WEB APPENDIX

A. Correlations Between Independent Variables

CORRELATION MATRIX

1 2 3 4 5 6 7 8 9 10 11 12

1. Trial dummy 1.00

2. Direct marketing .15*** 1.00

3. Advertising .08*** .04*** 1.00

4. Age .05*** .03*** .09*** 1.00

5. Household size .03*** .02*** .00 .10*** 1.00

6. Income .01 .04*** .24*** .11*** .04*** 1.00

7. Time-to-adoption .14*** .15*** .00 .01*** .07*** .02*** 1.00

8. Subscription Fee .22*** .09*** .03*** .02*** .03*** .00 .14*** 1.00

9. Cancellation penalty .36*** .11*** .00 .03*** .04*** .01** .26*** .76*** 1.00

10. VOD credit .22*** .07*** .02*** .00 .04*** .00 .27*** .70*** .62*** 1.00

11. Temperature .00 .08*** .07*** .00 .01 .01 .21*** .31*** .25*** .20*** 1.00

12. Time-since-adoption .08*** .07*** .03*** .02*** .04*** .01*** .29*** .71*** .79*** .56*** .17*** 1.00

Significance levels (two-tailed): *** p < .001, ** p < .01. The corresponding variance inflation factors (VIFs) do not exceed 10 in any of the models (Hair et al. 2010).

52

B. Parameter Estimates for the Propensity Model in the Matching Procedure

ESTIMATION RESULTS FOR THE PROPENSITY SCORE MODELa

Estimate (SE)

Intercept .315 (.011)

Age .005 (.000) Household size .027 (.003)

Income .072 (.004)

Time-to-adoption .009 (.001) Direct marketing .649 (.017)

Advertising .426 (.014)

Subscription fee .001 (.000)

Log-likelihood: -8,357; N: 16,512; BIC: 16,792; AIC: 16,731 a Numbers in bold are significant at the p < .05 level (two-sided).

C. Estimated Correlations Between Random Intercepts of Retention and Usage Models

ESTIMATED CORRELATIONS BETWEEN RANDOM

INTERCEPTSa

Correlation coefficient (SE)

1 2 3 4

1. Retention

2. UsageFR .083 (.011)

3. UsagePPU .117

(.005)

.033

(.006)

4. Zero-inflation model of UsagePPU

.058 (.028)

.279 (.011)

.728 (.011)

a Numbers in bold are significant at the p < .05 level (two-sided).

D. Computation of Net CLV

We compute a customer’s Net CLV as the discounted value of the expected future profits

minus the costs to acquire that customer:

(8) Net CLVi = ∑Sit×mit+(1−Sit)×Pit

(1+d)t36t=1 − ACi,

where Sit is the survival probability of customer i in month t after acquisition, mit is the

customer’s expected profit if he or she retains the service in month t, Pit is the penalty (in €) the

customer has to pay when he or she terminates the contract in month t (zero during a trial and

from month 12 onward), ACi refers to the customer’s acquisition costs, and d is the monthly

53

discount factor. d corresponds to an annual rate of 8.5%, which is a common value for the

weighted average cost of capital of TV content providers such as Netflix, Comcast, and DirecTV

(e.g., Cusick et al. 2014; Swinburne, Khadilkar, and Yeh 2014).

Notice that the survival probability Sit in month t is based on the customer’s retention

decisions in the preceding months: Sit = Prob(ri,0, …, ri,t-1 = 1). Furthermore, the expected margin

mit in a given month equals the revenue generated by the customer, less the costs to retain the

customer (Reinartz, Thomas, and Kumar 2005):

(9) mit = Feeitsub + FeePPU × UsagePPUit⏟

Revenue

− CostsDM × DMit − CostsAdv ⏟

Costs

,

where Feeitsub is the monthly subscription fee of €15.95 that customer i has to pay in month t

(zero during a free-trial period), FeePPU is the average per-unit fee for the pay-per-use service

corrected for average per-unit content cost (€3 minus €.25 throughout the simulation period1),

UsagePPUit refers to the customer’s expected consumption amount (in units) of the pay-per-use

service in month t, DMit is the number of direct-marketing contacts with the customer in month t,

CostsDM is the average cost for a single contact, and CostsAdv is the average monthly advertising

cost per customer.

To determine ACi, CostsDM , and CostsAdv , we need to allocate the observed marketing

efforts between acquisition and retention (Gupta, Lehmann, and Stuart 2004). Because we

observe consumer-specific direct-marketing efforts before and after acquisition, we can readily

compute the acquisition and retention direct-marketing costs for each customer by multiplying

1 To obtain an estimate of content costs for streamed VODs, we divide Netflix’s cost of revenue in 2013 ($2.6bn) by

the number of hours of watched content in that year (12bn), resulting in content costs of $.22 per streamed hour

(Netflix 2014). At an average length of 90 minutes per VOD (Huang, Li, and Ross 2007), content costs amount to

$.33 per VOD (approximately €.25).

54

the relevant number of direct-marketing contacts by the average delivery cost of €.37.2 Notice

that the average observed monthly number of marketing contacts for customer acquisition is .62

for free-trial, and .26 for regular customers. The company uses direct marketing less often for

customer retention: once acquired, free-trial and regular customers are contacted on average .34

and .16 times per month, respectively. In line with Gupta, Lehmann, and Stuart (2004), we use

these average retention efforts when simulating future behavior beyond the observation period.

We also need to split the expenditures on mass advertising into retention and acquisition

costs. We divide the monthly total advertising budget between retention and acquisition

proportionally to the numbers of iTV adopters and non-adopters in the company's total customer

base (see Gupta, Lehmann, and Stuart 2004). We then distribute the acquisition advertising

budget over the company’s new customers and the retention advertising budget over the already

acquired customers. On average, the company spends €169.61 on advertising to acquire a new

customer. This number is close to the US$181.8 reported in Libai, Muller, and Peres (2009) for

another telecommunication provider in the same geographical region. The monthly advertising

cost for retaining a customer is €.51, and thus substantially lower than the acquisition cost (see

Thomas 2001). Other fixed costs are not directly related to the number of customers or their

usage intensity (e.g., the network infrastructure which is owned by the company), and hence can

be ignored in the computation of Net CLV.

2 The number of iTV customers was small relative to the company’s entire customer base (which mainly includes

telephone and Internet subscribers), so that we can safely assume that the already existing service employees

handled iTV customers as well. As a result, direct-marketing costs correspond to the actual delivery costs of the

messages. There are three channels for direct marketing in the data set: postal mail, outbound calls, and e-mail. We

take actual postal fees of the National Post in 2004 (€.44 for non-priority delivery), contact time of the average

phone call (6.39 minutes for non-technical inquiries; Sun and Li 2011) multiplied by the per-minute call fee of the

national carrier in 2004 (€.051), and a nominal amount of €.01 to reflect the small delivery costs of emails (Danaher

and Dagger 2013). Finally, we weight the delivery costs by the relative occurrence of the respective channels in the

data set and arrive at an average direct-marketing cost of €.37 per contact.

55

References

Cusick, Philip, Richard Choe, Eric Pan, and Ava Zhang (2014), “DirecTV: Takeways from DTV

Management Meetings - US and Latam [J.P. Morgan Research],” Thomson ONE Banker

Analytics (accessed 30 September 2014), http://banker.thomsonib.com/

Danaher, Peter J. and Tracey S. Dagger (2013), “Comparing the Relative Effectiveness of

Advertising Channels: a Case Study of a Multimedia Blitz Campaign,” Journal of

Marketing Research, 50 (4), 517-34.

Gupta, Sunil, Donald R. Lehmann, and Jennifer Ames Stuart (2004), “Valuing Customers,”

Journal of Marketing Research, 41 (1), 7-18.

Hair, Joseph F., William C. Black, Barry J. Babin, and Rolph E. Anderson (2010), Multivariate

Data Analysis (7th ed.). Upper Saddle River, New Jersey: Pearson.

Huang, Cheng, Jin Li and Keith W. Ross (2007), “Can Internet Video-on-Demand be

Profitable?,” Proceedings of ACM SIGCOMM 2007.

Kumar, V., Rajkumar Venkatesan, Tim Bohling, and Denise Beckmann (2008), “Practice Prize

Report—The Power of CLV: Managing Customer Lifetime Value at IBM,” Marketing

Science, 27 (4), 585-99.

Libai, Barak, Eitan Muller, and Renana Peres (2009), “The Diffusion of Services,” Journal of

Marketing Research, 46 (2), 163-75.

Netflix, Inc. (2014). Form 10-K 2013 (Annual Report).

Reinartz, Werner J. and V. Kumar (2003), “The Impact of Customer Relationship Characteristics

on Profitable Lifetime Duration,” Journal of Marketing, 67 (1), 77-99.

Reinartz, Werner J., J. S. Thomas, and V. Kumar (2005), “Balancing Acquisition and Retention

Resources to Maximize Customer Profitability,” Journal of Marketing, 69 (1), 63-79.

56

Swinburne, Benjamin, Hersch Khadilkar and Thomas Yeh (2014), “Netflix Inc: Europe Gives,

and Europe Takes Away [Morgan Stanley Research],” Thomson ONE Banker Analytics

(accessed 30 September 2014), http://banker.thomsonib.com/

Sun, Baohong and Shibo Li (2011), “Learning and Acting on Customer Information: A

Simulation-based Demonstration on Service Allocations with Offshore Centers,” Journal

of Marketing Research, 48 (1), 72-86.

Thomas, Jacquelyn S. (2001), “A Methodology for Linking Customer Acquisition to Customer

Retention,” Journal of Marketing Research, 38 (2), 262-68.

Venkatesan, Rajkumar and V. Kumar (2004), “A Customer Lifetime Value Framework for

Customer Selection and Resource Allocation Strategy,” Journal of Marketing, 68 (4),

106-25.